app.py

import torch
from transformers import DistilBertTokenizer, DistilBertForSequenceClassification
import gradio as gr
import re
import nltk
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer
from nltk.tokenize import word_tokenize

# ======== Download NLTK Resources ========
nltk.download('stopwords')
nltk.download('punkt_tab')
nltk.download('wordnet')
nltk.download('omw-1.4')

# ======== Preprocessing Setup ========
stop_words = set(stopwords.words('english'))
lemmatizer = WordNetLemmatizer()

def preprocess_text(text):
    # Remove non-alphabetic characters
    text = re.sub(r'[^A-Za-z\s]', '', text)
    # Remove URLs
    text = re.sub(r'http\S+|www\S+|https\S+', '', text)
    # Normalize spaces
    text = re.sub(r'\s+', ' ', text).strip()
    # Lowercase
    text = text.lower()
    # Tokenize
    tokens = word_tokenize(text)
    # Remove stopwords
    tokens = [word for word in tokens if word not in stop_words]
    # Lemmatize
    tokens = [lemmatizer.lemmatize(word) for word in tokens]
    return ' '.join(tokens)

# ======== Class Names ========
class_names = [
    "Anxiety",
    "Bipolar",
    "Depression",
    "Normal",
    "Personality disorder",
    "Stress",
    "Suicidal"
]

# ======== Load Tokenizer & Model ========
tokenizer = DistilBertTokenizer.from_pretrained("distilbert-base-uncased")
model = DistilBertForSequenceClassification.from_pretrained(
    "distilbert-base-uncased", 
    num_labels=len(class_names)  # 7 classes
)
model.load_state_dict(torch.load("best_model.pth", map_location=torch.device("cpu")))
model.eval()

# ======== Prediction Function ========
def predict_text(text):
    cleaned_text = preprocess_text(text)
    if not cleaned_text.strip():
        return {cls: 0.0 for cls in class_names}

    inputs = tokenizer(cleaned_text, truncation=True, padding=True, max_length=128, return_tensors='pt')
    with torch.no_grad():
        outputs = model(**inputs)
        probs = torch.softmax(outputs.logits, dim=1).flatten().tolist()

    return {cls: float(prob) for cls, prob in zip(class_names, probs)}

# ======== Gradio Interface ========
demo = gr.Interface(
    fn=predict_text,
    inputs=gr.Textbox(lines=4, placeholder="Enter your statement here..."),
    outputs=gr.Label(num_top_classes=len(class_names)),
    title="Mental Health Sentiment Classifier",
    description="Classifies text into mental health categories: Normal, Depression, Suicidal, Anxiety, Bipolar, Stress, Personality disorder."
)

if __name__ == "__main__":
    demo.launch()